Unit Roots & ADF Test

ECON 3209 · Week 19, Lecture 2 · Kerala Agricultural University

Department of Development Economics, KAU

Autumn 2026

Learning Outcomes

By the end of this lecture, you should be able to:

  1. define a unit root and relate it to random-walk behaviour
  2. run and interpret the Augmented Dickey-Fuller test
  3. compare ADF and KPSS hypotheses
  4. test level and differenced series in Python
  5. apply these ideas to Kerala paddy production data

What Is a Unit Root?

  • A simple random walk is:

\[Y_t = Y_{t-1} + \varepsilon_t\]

  • Shocks have permanent effects.
  • The series wanders without returning to a fixed mean.
  • This is the classic nonstationary unit-root case.

ADF Test Logic

  • The Augmented Dickey-Fuller test estimates a regression with lagged levels and lagged differences.
  • Null hypothesis: the series has a unit root (nonstationary).
  • Alternative: the series is stationary.
  • A small p-value leads us to reject the unit-root null.

KPSS Complements ADF

  • ADF null: unit root.
  • KPSS null: stationarity.
  • Because the nulls differ, the two tests complement each other.
  • Together they provide a more balanced diagnosis.

Simulating a Stationary Series and a Random Walk

ADF and KPSS in Python

Reading the Tests

  • ADF small p-value → evidence against a unit root.
  • KPSS small p-value → evidence against stationarity.
  • Strongest conclusion comes when the two tests agree.
  • If they disagree, inspect plots, sample size, lag choice, and economic context.

Kerala Paddy Series

Visualising Level and First Difference

Decision Grid

ADF result KPSS result Typical interpretation
Reject unit root Fail to reject stationarity Stationary
Fail to reject unit root Reject stationarity Nonstationary
Both reject Possible structural break / ambiguity
Both fail to reject Low power / inconclusive

Always combine tests with economic reasoning.

Why It Matters for Forecasting

  • ARIMA models require correct differencing.
  • If we ignore unit roots, forecasts may be unstable and inference misleading.
  • Proper testing helps choose whether the integration order \(d\) should be 0, 1, or higher.

Exercise

Run ADF and KPSS on the paddy series in levels and first differences. Which version looks more stationary, and what does that imply for ARIMA differencing?

Summary

  • ✅ A unit root implies shocks have persistent effects and the series is usually nonstationary in levels.
  • ✅ ADF tests the null of a unit root, while KPSS tests the null of stationarity.
  • ✅ Testing both levels and first differences helps determine the amount of differencing needed.
  • ✅ This decision is central for building ARIMA models in the next week.

Next Lecture

  • We use ACF and PACF plots to identify AR and MA structure.
  • You will learn how lag patterns guide order selection.
  • This is the bridge from stationarity testing to ARIMA design.